sec. 10.

1 MINIMIZATION OF FUNCTIONALS 271

~
1
The basic problem of the calculus of variations may be illustrated by the
following example. Let us consider a functional J[u] defined by the integral
10
(1) J[u] =f F(x , u , u') dx .

We shall give our attention to ali functions u(x) which are continuous and
VARIATIONAL CALCULUS, differentiable, with continuous derivatives u'(x) and u"(x) in the interval
a < x < b, and satisfying the boundary conditions
ENERGY THEOREMS, (2) u(a) = Uo , u(b) = Ui,

SAINT-VENANT'S PRINCIPLE where u0 and u1 are given numbers. We assume that the function F(x, u, u')
in Eq. (1) is continuous and differentiable with respect to x, and all such u,
and u', up to all second-order partial derivatives, which are themselves
continuous.
There are at least three important reasons for taking up the calculus of Among all functions u(x) satisfying these continuity conditions and
variations in the study of continuum mechanics. boundary values, we try to find a special one u(x) = y(x), with the property
l. Because basic mínimum principies exist, which are among the most that J[u] attains a mínimum when u(x) = y(x), with respect to a sufficiently
beautiful of theoretical physics. small neighborhood of y(x). The neighborhood (h) of y(x) is defined as
2. The field equations (ordinary or partial differential equations) and the follows. lf h is a positive quantity, a function u(x) is said to lie in the neighbor-
associated boundary conditions of man y problems can be derived from hood (h) of y(x) if the inequality
variational principies. In formulating an approximate theory, the
shortest and clearest derivation is usually obtained through variational (3) Jy(x) - u(x)I <h
calculus. holds for all x in (a, b). The situation is illustrated in Fig. 10.1 :l.
3. The "direct" method of solution of variational problems is one of the
most powerful tools for obtaining numerical results in practica! u
problems of engineering importance.
In this chapter we shall discuss several variational principies and their uor---- ,
1
applications. 1
1

A brief introduction to the calculus of variations is furnished below.

1
1 y(x)
1
1
Those readers who are familiar with the mathematical techniques of the 1
1
b X
calculus of variations may skip over the first six sections. a

u,~------------------------~~~~-- '
10.1. MINIMIZATION OF FUNCTIONALS Fig. 10.1 :l. Functions u(x) and y(x) .

The calculus of variations is concerned with the minimization of functionals . Let us assume that the problem posed above has a solution, which will be
rf u(x) is a function of x, defined for x in the interval (a, b), and if / is a designated by y(x); i.e., there exists a function y(x) such that the inequality
quantity defined by the integral
(4) <
f
J[y] J[u]
2
I = [u(x)] dx, holds for all functions u(x) in a sufficient!y small neighborhood (h) of y(x).
Let us exploit the necessary consequences of this assumption.
then the value of / depends on the function u(x) as a whole. We may indicate Let 11(x) be an arbitrary function with the properties that 11(x) and its
this dependence by writing J[u(x)]. Such a quantity / is said to be afunctional derivatives r¡'(x), 11"(x) are continuous in the interval a < x < b, and that
of u(x). Physical examples of functionals are the total kinetic energy of a
(5) r¡(a) = r¡(b) = O.
fl.ow field and the strain energy of an elastic body.
270
 272 ENERGY THEOREMS Chap. 10 Sec. 10.1 MINIMIZATION OF FUNCTIONALS 273

Then the function . holds Jor ali Junctions r¡(x) which vanish at x = a and b and are continuous
1
(6) u(x) = y(x) + €r¡(x) together with theirftrst 2n derivatives, where nis apositive integer, then '1/'(X) O. =
satisfies all the continuity conditions and boundary values specified at the Proof This lemma is easily proved indirectly. We shall show first that
beginning of this section. In fact, any function u(x) satisfying these conditions '1/'(x) =O in the open interval a < x < b. Let us suppose that this statement
can be represented with sorne function r¡(x) in this manner. For a su$ciently is not true, that '1/'(x) is different from zero, say positive, at x = ~. where ~
small c5 > O, this function u(x) belongs, for ali€ with /€/ < c5, to a prescribed Jies in the open interval. Then, according to the continuity of '1/'(x), there
neighborhood (h) of y(x). Now we introduce the function must existan interval ~ - c5 < x < ~ + c5 (with a < ~ - c5, ~ + c5 < b, c5 >
O), in which '1/'(x) is positive. Now we take tbe function
(7) cp(~) = J[y + €TJ] = f F[x, y(x) + €r¡(x), y'(x) + €TJ'(x)] dx. (x - ~ + c5) n(x -
4
~ - c5) 4n in~ - c5 < x < ~ + c5,
Since y(x) is assumed to be known, cp(€) is a function of € for any specific (14) r¡(x) = {
O elsewhere,
r¡(x). According to (4), the inequality
(8) cp(O) < cp( €) where nis a positive integer. (See Fig. 10. l :2.) This function r¡(x) satisfies the
continuity and boundary conditions
must hold for all € with /€/ < c5. In other words, cp(€) attains a minimum at specified above. But the choice of
€ = O. The function cp(€) is differentiable with respect to €. Therefore, the
necessary condition for cp(€) to attain a minimum at € = O must follow,
l9) cp'(O) = O,
where a prime indicates a differentiation with respect to €. Now, a differentia-
r¡(x) as defined by (14) will make

f '1/'(x)r¡(x) dx > O, in contradiction

to the hypotbesis. Thus, the hypoth-
esis is untenable, and '1/'(x) O in =
a
Fig. 10.1
-o t +o
:2. The
lfunction
b
r¡(x).
~' ~X

tion under the sign of integration yields

a < x < b. According to the conti-
(10) nuity of '1/'(x), we get '1/'(x) = O also in a < x < b. Q.E.D. We remark that
cp'(€) = f [Fu(x, Y+ €TJ, y'+ €TJ )r¡(x)
1
+ F ,,.(x, y+ €TJ, y'+ €TJ')r¡'(x)] dx. the lemma can be extended to hold equally well for multiple integrals.

where F,,, F,,. indicates oF/ou, oF/ou', respectively. Integrating the last From Eq. (12), it follows immediately from the lemma that the factor in
integral by parts, we obtain front of r¡(x) must vanish:
(11)
(15a) • Fv(x, y, y') - .!!:_ F 11 ,(x, y, y') = O, a < x < b.
1 dx
cp'(€) = f [F,,(x, Y+ €TJ, y' + €TJ ) - ; F ,,.(x, y + €TJ , y'+ €TJ')}(x) dx
Tbis is a differential equation that y(x) must satisfy and is known as Euler's
equation. Written out in extenso, we have
+ F ,,.(x, y + €TJ, y' + €TJ')r¡(x) 1:
d 2y aF2
, dy aF 2
,
The last term vanishes according to (5). cp'(€) must vanish at € = O, at which
u equals y. Hence, we obtain the equation
(15b) • dx 2 oy' oy' (x, y, y) + dx oy' oy (x, y, y)
+ -aF- (x, y, y') -
2
oF
- (x, y, y') = O.
(12) O= cp'(O) = f [ Fv(x, y, y') - ; F,,A x, y, y')}(x) dx, oy'ox oy

which must be valid for an arbitrary function r¡(x). Should the problem be changed to finding the necessary condition for
J[u] to attain a maximum with respect to a sufficiently small neighborhood
Equation (12) leads immediately to Euler's differential equation, by
(h) of y(x), the same result would be obtained. Hence, the result: The validity
virtue of the following "fundamental lemma of the calculus of variations :"
of Euler's differential Eq. (15) is a necessary condition Jor a Junction y(x) to
Jurnish an extremum of the Junctional J[u] with respect to a sufficiently small
LEMMA. Let '1/'(x) be a continuous Junction in a < x < b. 1J the relation
neighborhood (h) of y(x).
(13) f '1/'(x)r¡(x) dx = O The satisfaction of Euler's equation is a necessary condition for J[u] to
attain an extremum; but it is not a sufficient condition. The question of
274 ENERGY THEOREMS Chap. 10 Sec. 10.2 FUNCTIONAL INVOLVING HIGHER DERIVATIVES 275

sufficiency is rather involved ; an interested reader must refer to treatis'j::s on Example 3. Minimum surface of revolution
calculus of variations such as those listed in Bib. 10.1, on p. 494.i" ·
Now a point of notation. It is customary to call er¡(x) the variation of Find y(x) that minimizes the functional
u(x) and write
(16) er¡(x) = ou(x).
J[u] = 21r lb uV1 + u' 2 dx, u(a) = A , u(b) = B.

Jt is also customary to define the first variation of the functional J[u] as Ans. The Euler equation can be integrated to give
(17) OJ = E </>'( E). ,2

On multip'Jying both sides of Eq. (11) by e, it is evident that it can be written as yVI + y' 2 - / yy = C1.
. 1 + y'2

(18) oJ = f [Fu(x, u, u') - :X F u,(x, u, u')] ou(x) dx Let y' = sinh t, then y = c1 cosh t and
dy c 1 sinh t dt
dx = - = = e1 dt X = C1f + C2.
+ F u·(x, u, u') ou(x) 1:. y' sinh t '

Hence, the minimum surface is obtained by revolving a curve with the following
This is analogous to the notation of the differential calculus, in which the
parametric equations about the x-axis,
expression df = ef'(x) for an arbitrary small parameter e is ca!led the "dif- X - c
ferential of the function/(x)." It is obvious that oJ depends on the function X = C1f + C2, y = c1 cosh t, or y = c1 cosh - -2 ,
C1
u(x) and its variation ou(x). Thus, a necessary condition for J[u] to attain an
extremum when u(x) = y(x) is the vanishing of the first variation oJ for all which is a family of catenaries. The constants c1 , c2 can be determined from the "
variations ou with ou(a) = ou(b) = O. end points.

Example J.

![u] = L\1 + u' 2) dx = min, u(a) = O, u(b) = l.

10.2. FUNCTIONAL INVOLVING HIGHER DERIVATIVES
OF THE DEPENDENT VARIABLE
The Euler equation is
d d In an analogous manner one can treat the variational problem connected
- dxF11, = dx2y' = 2f = O. with the functional
Hence y(x) is a straight line passing through the points (a, O) and (b, 1). (1) J[u] = ibF(x, u, u( 1 >, . . . , u(n>) dx
Example 2. involving u(x) and its successive derivatives un>(x), u( 2 l(x), ... , u< " >(x), for
a < x < b. To state the problem concisely we denote by D the set of all real
Which curve minimizes the following functional? functions u(x) with the following properties:
{'r/2
I[u] = Jo [(u') 2 - u2 ] dx, u(O) = O, u(i1r) = l. (2a) u(x), ... , u< 2">(x) continuous in a < x < b,
Ans. y = sin x. (2b) u(a) = oto, u<•>ca) = ot., y = l, ... ' n - l,
t Any consideration of the sufficient conditions requires the concept of the second (2c) u(b) = /Jo, u<•>(b) = {J., v = l, ... , n - l,
variations and the examination of its positive or negative definiteness. Severa! conditions
are known; they are similar to, but more complex than, the corresponding conditions for where ot 0 , /Jo, ot, and /J. are given numbers. A function u(x) which possesses
maxima or minima of ordinary functions . It is useful to remember that many subtleties these continuity properties and boundary values is said to be in the set D.
exist in the calculus of variations. A physicist or an engineer rarely worries about the We assume that the function F(x, u, un>, ... , u<">) has continuous partial
mathematical details. His minimum principies are established on physical grounds and the
derivatives up to the order n + 1 with respect to the n + 2 arguments
existence of a solution is usually taken for granted. Mathematically, however, many
examples can be constructed to show that a functional may not have a maximum or a x, u, un>, . .. , u< n> for u in the set D. We seek a function u(x) = y(x) in D
minimum, or that a solution of the Euler equation may not minimize the functional. An so thatJ[u] is a mínimum (ora maximum) for u(x) = y(x), with respect to a
engineer should be aware of these possibilities. sufficiently small neighborhood (h) of y(x).
 276 ENERGY THEOREMS Chap. 10 Sec. 10.3 SEVERAL UNK.NOWN FUNCTIONS 277

Let r¡(x) be an arbitrary function with the properties l. be a functional depending on m functions u1(x), ... , um(x). We assume that
. ·' the functions F(x, u1, ... , u;,.), u1(x), ... , um(x) are twice differentiable,
r¡(x), r¡<ll(x), ... , r¡< 2n>(x) continuous in a < x < b, . and that the boundary values of u1, ... , um are given at x = a and b. We
(3)
r¡(a) = ... = r¡<n-1>(a) = r¡(b) = ... = r¡<n-l>(b) = O. seek a special set of functions uµ(x) = yµ(x), µ = I, ... , m, in order that
J[u 1 , ••• , um] attains a mínimum (ora maximum) when uµ(x) = yµ(x), with
Then the function u(x) = y(x) +
Er¡(x) belongs to the set D for ali E. For respect to a sufficiently small neighborhood of the y µ(x); i.e., for all uµ(x)
sufficiently small E's this function belongs to a prescribed neighborhood (h) satisfying the relation
of y(x). Once again, we introduce the function
lyµ(x) - uµ(x)I < hµ, hµ > O, µ = 1, 2, ... , m.
. rb
(4) cp(E) = J[y + Er¡] = Ja F(x, Y+ ErJ, y< 1> + Er¡<ll, ... , y<n> + Er¡<n>) dx. Again it is easy to obtain the necessary conditions. Let the set offunctions
yi(x), .. . ,Ym(x)be a solution ofthevariational problem. Let r¡ 1(x), ... , r¡m(x)
The assumption that y(x) minimizes J[u] leads to the necessary condition be an arbitrary set of functions with the properties
<p'(O) = O. So we obtain by differentiating under the sign of integration,
integrating by parts, and using Eqs. (3), the result r¡µ(x), r¡ix), r¡;(x) continuous in a < x < b,
(2)
r¡µ(a) = r¡µ(b) = O, µ = I, ... , m.
O= <p'(O) =
Ja
! [oyºe,> F(x, y, y<1>, ... , y<n>)]r¡<•>(x)} dx
(b {
v=O
Toen consider the function
(5)
= (b {!(-1)' d'J ºe,> F(x, y, y< 1l, ... , y<n>)J }r¡(x) dx, (3) <p(E1, · · • , Em) + E1rJ1, • • · , Ym + EmrJm]
= J[y¡
Ja v=O dx .._oy
which holds for an arbitrary function r¡(x) satisfying (3). According to the
f
= F(x, Y1 + E1rJ1, · · ·, Ym + EmrJm; y{+ E¡r¡{, ... , y;,.+ Emr¡;,.) dx.
lemma proved in Sec. 10.1, we obtain the Euler equation Since yi(x), ... , Ym(x) minimize (or maximize) J[y + ErJi, ... , Ym + E,,.rJm],
1
the following inequality must hold for sufficiently small E1, ... , Em:
(6) ... ~(
,C., -
o F( x, y, y <1> ' ... 'y <n>)]
l)' -d'. [ ~ = O'
v=O dx oy (4) <p( €¡, ... , Em) )> <p(O, ..• , O), [or <p(Ei, ... , Em) < cp(O, ... , O)].
which is a necessary condition for the minimizing ( or maximizing) function The corresponding necessary conditions are
y(x).
The notations bu, IJJ, etc., introduced in Sec. 10.1, can be extended to this
case by obvious changes.
(5) ª<p) =
( 0€µ o, for €¡ = €2 = .. . = Em = O, µ= 1, .. . ,m,
which lead to
Example

Find the extremal of J[u] = i 1

(1 + u"2) dx satisfying the boundary conditions
(6) 0= ib [F11 µ(x, Y1, . · · , Ym, y{, · · . , y;,.)r¡µ(x)
+ F )x, Y1, ... , Ym, y{, ... , y;,.)r¡~(x)] dx,
11
u(O) = O, u'(O) = 1, u(l) = 1, u'(l) = l. µ= 1, ... , m.
Ans. y= x. Using integration by parts and the conditions (2), we obtain

10.3. SEVERAL UNKNOWN FUNCTIONS (7) O=

Ja(b [F11 µ (x, Y1, . · · , Ym, Y~, . · . , y;,.)
The method used in the previous sections can be extended to more com- - !!:_ F 11 , (x, Yi, ... , Ym, y{, .. . , y;,.)]r¡µ(x) dx
dx µ
plicated functionals. For example, let
+ F11' µ(x, Yi, · · ·, Ym, Y~, ... , y;,.)r¡µ(x) lb,
(1) J[u 1 , u 2, .•• , um] = f F(x, u 1 , .•. , um; u~, ... , u;,.) dx
a
µ= 1, ... ,m.
278 ENERGY THEOREMS Chap. 10 Sec. 10.4 279
SEVERAL INDEPENDENT VARIABLES

Equation (7) must hold for any set of functions r¡(x) satisfying (2).. By f h~ five arguments. Let D be the set of all functions u(x, y) with the following
lemma of Sec. 10.1 we have the following Euler equations which must be properties:
satisfied by y 1 (x), ... , Ym(x) minimizing (or maximizing) the functional (1): (a) u(x, y), u.,(x, y), uv(x,_Y), u~(x, y),
(8) ~ Fv/x, Yi, · · ·, Ym; Y~,· ··,y;,.) (2) D: u"'1/(x,y), u11v(x,y) contmuous m G,
{
- !!:_ Fv· (x, Y1, ... , Ym; Y1: . , . , y,:.) = O, µ = l, ... ,m. (b) u(x,y) prescribed on C.
dx µ
We now seek a special function u(x, y) = v(x, y) in the set D, which minünizes
As a ge~eralization of the variational notations given in Sec. 10.1 the (or maximizes) the functional J[u].
expression Let v(x, y) be a solution of this variational problem. Let r¡(x, y) denote an
(12) CJJ = i: Eµ(ªc/>)
µ= l ªEµ
arbitrary function with the properties
(a) r¡(x, y) has continuous derivatives up to the
is called the "first variation of the functional." (3) second order in G,
Variational problems for functionals involving higher derivatives of {
(b) r¡(-', y) = O for (x, y) on the boundary C.
ui, ... , um can be treated in the same way.
A consideration of the function
Example (4) cf,(E) = J[v + ErJ],
Find the extremals of the functional which attains an extremum when E = O, again leads to the necessary condition
f"/2
J[y, z] = Jo (y' 2 + z' 2 + 2yz) dx, (5) dcf, (O)= O,
dE
i.e., explicitly,
y(O) O,
= y(i) 1, = z(O) O, z(i)
= -1.

The Euler equations are (6) O= JL{Fv(x, y, v, v.,, v11 )r¡ + Fv.(x, y, v, v.,, vv)r¡.,
y" - z = O, z" - y = O.
+ Fv,(x, y, v, v.,, v )r¡iJ dx dy.
11

Eliminating z, we have yiv - y = O. Hence, The last two terms can be simplified by Gauss' theorem after rewriting (6) ast

Z =
J =
y·
C¡e"'

= C¡e"'
+ C2e-"' + C3 cos X + C4 sin x,
+ C2e-"' - C3 cos X - C4 sin X. o= fL{Fv. r¡ + :X (Fv•. r¡) - r¡ a~ Fv. + ! (Fv•. r¡) .- r¡ ! Fv.} dx dy.

From the boundary conditions we obtain the solution An application of Gauss' theorem to the sum of the second and fourth terms
in the integrand gives
y-= sin x, z = -sin x.

(7) O =J.·fJo {Fv - axaFv• - ayaFv•)r¡(x, y) dx dy

10.4. SEVERAL INDEPENDENT VARIABLES
+Jef {Fv• · ni(s) + Fv• · nls)}r¡ ds
Consider the functional

(1) J[u] = fLF(x, y, u, u.,, u11 ) dx dy,

Here n1 (s) and nls) are the components of the unit outer-normal vector
n(s) of C.
A

where G is a finite, closed domain of the x, y-plane with a boundary curve C

t The symbol (íJ/íJx)F•• means that F•• should be considered as a function of x and y,
e.g.,
which has a piecewise continuously turning tangent. The function F(x, y, u,
u.,, u11 ) is assumed to be twice continuously differentiable with respect to its -F.
a =- F.+
íJ
--+
íJF•• ov
- - +ov,-íJx
-·
íJF•• ov. oF•• ov,
íJx •• íJx •• ov íJx ov. íJx
280 ENERGY THEOREMS Chap. 10 Sec. 10.5 SUBSIDIARY CONDITIONS-LAGRANGIAN MULTIPLIERS 281

y
The line integral in (7) vanishes according to (3). The function :17(x, )') in A
the surface integral is arbitrary. The generalized lemma of Sec. ,10.1 'then
leads to the Euler equations

F --F
V ax
a v.
--F
a
ay v,
=0·,
i.e.,
oF é)2p é)2p o2F ov o2F ov
(9) ..
. •.OV ov., OX ovll ay ov., ov ox ovll ov ay
é)2F o2v é)2F o2v é)2F o2v -+---.....---------,.....-----x
---- 2 2-------- 0
ov! OX ov., ovll ox ay ov! 0y 2 •

Equation (9) is a necessary condition for a function v(x, y) in the set D
minimizing (or maximizing) the functional (1).
In the same way one may treat variational problems connected with
functionals which involve more than two independent variables, higher
derivatives, and several unknown functions.
Example 1
Fig. 10.5:1. Illustrating a subsidiary condition.
that lie on the segment of the curve AB. For conciseness of expression, we
shall call such a subdomain-the segment AB-G<P.
Let us find the necessary conditions for f(x, y) to attain an extreme value
at a point (x, y) in G</>' with respect to ali points of a suffi.ciently small neigh-
borhood of (x, y) belonging to G<P.

J[u] = JL (u; + u!) dx dy,

Let us assume that the function cp(x, y) has continuous partial derivatives
with respect to x and y, and that at the point (x, y) not both derivatives are
zero; say,
with a boundary condition that u is equal to an assigned function f(x, y) on the
boundary C of the domain D. The Euler equation is the Laplace equation (2) cpy(x, y) =1- O.
Va;a; + V 1111 = 0. Then, according to a fundamental theorem on implicit functions, there exists
a neighborhood of x, say x - b < x < x + b (b > O), where the equation
Example 2
JL
cp(x, y) = O can be solved uniquely in the forro
J[u] = (u;., + u!11 + 2u.,11 )2 dx dy = min. (3) y = g(x).
Then v must satisfy the biharmonic equation The function g(x) so defined is single-valued and differentiable, and
o4v o4v o4v cp[x, g(x)] = O is an identity in x . Hence
ox4 + 2 ax2ay2 + oy4 = o.
(4) O = cp.,[x, g(x)] + cp [x, g(x)] dg(x). '
11
dx
10.5. SUBSIDIARY CONDITIONS-LAGRANGIAN Now, let us consider the problem of the extremum value of f(x, y) in G4,.
According to (3), in a suffi.ciently small neighborhood of (x, y), y is an
MULTIPLIERS
implicit function of x, and f(x, y) becomes
In many problems, we are interested in the extremum of a function or (5) ff(x) = f[x, g(x)].
functional under certain subsidiary conditions. As an elementary example, If ff (x) attains an extreme value at x, then the first derivative of ff(x)
let us considera function/(x,y) defined for all (x, y) in a certain domain G. vanishes at x; i.e.,
Suppose that we are interested in the extremum off(x, y), not for ali points in
G, but only for those points in G which satisfy the relation (6) o + fu(x, y) dg (x).
= dff (x) = J.,(x, y)
dx dx
(1) cp(x, y) = O. But the derivative dg/dx can be eliminated between (6) and (4). The result,
Thus, if the domain G and the curve cp(x, y) = O are as shown in Fig. 10.5 :1, together with (1), constitutes the necessary condition for an extremum of
then we are interested in finding the extremum of f(x, y) among all points f(x,y) in G4,.
282 ENERGY THEOREMS Chap. 10 Sec. 10.6 NATURAL BOUNDARY CONDITIONS 283

The formalism will be more elegant by introducing a number A, cal!ed two given points on this surface (Fig. 10.5 :2). Here we must minimize
Lagrange's multiplier, defined by
( 7) A = _ filx, y) .
cpv(x, y)
·' the functional
(13) l=
"f ''.J. +1 y' 2 + z' 2
dx,
"'º
A combination of (4), (6), and (7) gives with y(x), z(x) satisfying the condition cp(x, y, z) = O. This problem was
solved in 1697 by Johann Bernoulli, but a general method of solution was
(8) f,ix, y) + Acp.,(x, y) = o, given by L. Euler and J. Lagrange.
while (7) 1!1ªY be written as We consider a new functional

f"''
(9) '· fv(x, y) + Acpv(x, y) = o.
Equations (1), (8), and (9) are necessary conditions for the functionf(x,y) to
(14) J*[y, z, A]= [)1 + y' + z' + A(x)cp(x. y, z)] dx
2 2

"'º
attain an extreme at a point (x, y), where cp; + cp; > O. These conditions and use the method of Sec. 10.3 to determine the functions y(x), z(x), and
constitute three equations for the three "unknowns" x, y, and A. A(x) that minimizes I*. The necessary conditions are
These results can be summarized in the following manner. We introduce
a new function (15) ). ocp - .É_ y' = O,
(10) F(x,y; A)= J(x,y) + Acp(x,y). oy dx)l + y' + z'
2 2

If the functionf(x, y) has an extreme value at the point (x, y) with respect to (16) A ocp - !!:._ z' = O,
G,t,, and if oz dx )1 + y' + z'
2 2

(11) [cp.,(x, y)] 2 + [cp,,(x, y)]2 > o, (17) cp(x, y, z) = O.

then there exists a certain number .X so that the three partial derivatives of This system of equations determines the functions y(x), z(x), and A(x).
F(x, y; A) with respect to x, y, and A are zero at (x, y, l):

oF (x, y, 1) = f.,(x, y) + .Xcp.,(x, y) = o, 10.6. NATURAL BOUNDARY CONDITIONS

ox
In previous sections we considered variational problems in which the
(12) oF (x, y, l) = f,ix, y) + .Xcp/x, y) = o, admissible functions have prescribed values on the boundary. We shall now
oy
consider problems in which no boundary values are prescribed for the
oF (x, y, 1) = cp(x, y) = o. admissible functions. Such problems lead to the natural boundary conditions.
oA Consider again the functional ( 10.1 : 1) but now omit the boundary
If there exist points (x', y') in G with cp(x', y') = O and cp.,(x', y') = condition (10.1 :2). Following the arguments of Sec. 10.1, we obtain the
cpv(x', y') = O, additional considerations are necessary. necessary condition (10.l :11),

z
In the formulation (10) and (12),
the theorem can be generalized to the
case of n variables x1 , . • . , xn and
(1) O= f {F,, - d~ F 11 }i(x) dx +F 11 , • 17(x) 1:-

severa! subsidiary conditions O = cp¡(x1 , This equation must hold for all arbitrary functions 17(x) and, in particular, for
... , Xn) = ... = 'Pm(X1, ... , Xn), m < n. arbitrary functions 17(x) with ri(a) = ri(b) = O. This leads at once to the
The application of the Lagrange Euler equation (10.1 :15)
multiplier method to the rninimization of
(2) F,ix, y, y') - .!!_ F 11.(x, y, y') = O, in a < x < b.
---
--f'I
,1,1
¡11
functionals follows a similar reasoning. dx
As an example, let us consider the
gJ .,! classical problem of geodesics: to find
In contrast to Sec. 10.1, however, the last term F 11 , • ri(x) does not vanish J~,
X by prescription. Hence, by Eqs. (1) and (2), we must have
the line of rninimal length lying on a
Fig. 10.5:2. A problem of geodesics. given surface cp(x, y, z) = O and joining (3) (F11 , • ri):i:=b - (F11 , • 17).,=a = O
 284 ENERGY THEOREMS Chap. 10 Sec. 10.7 THEOREM OF MINIMUM POTENTIAL ENERGY 285

for all functions r¡(x). But now r¡(a) and r¡(b) are arbitrary. Taking, swo of such an order of magnitude that the material remains elastic. Such
functions r¡ 1 (x) and r¡ 2(x) with arbitrary displacements '5u; are called virtual displacements.
Let us assume that static equilibrium prevails and compute the "virtual
(4) 'Y/1(a) = 1, 'Y/1(b) = O; r¡ia) = O, r¡ib) = · 1;
•
we get, from (3), work" done by the body force F; per unit volume and the surface force T;*
per unit area:
(5) F11,[a, y (a), y'(a)] = O, F11,[b, y(b), y'(b)] = O.
The conditions (5) are called the natural boundary conditions of our problem.
V
LF;'5u; dv + fst:'5u; dS.
They are the boundary conditions which must be satisfied by the function On substituting T;* = <f;;V; and transforming according to Gauss' theorem,
y(x) if the functiorial J[u] reaches an extremum at u(x) = y(x), provided that we have
y (a) and y(b) are entirely arbitrary.
Thus, if the first variation of a functional J[u] vanishes at u(x) = y(x), (1) r t: '5u; dS = Jsr .
Js
<1; ; '5u;V; dS
and if the boundary values of u(x) at x = a and x = b are arbitrary, then
y(x ) must satisfy not only the Euler equation but also the natural boundary = L(<1;; '5u;),; dv
conditions which, in general, involve the derivatives of y(x). In contrast to
the natural boundary conditions, the conditions u(a) = ex, u(b) = {J, which
specify the boundary values of u(x) at x = a and b, are called rigid boundary
= f v <1;; ,; '5u; dv + fv <1; 1'5u;,; dv.
conditions. According to the equation of equilibrium the first integral on the right-hand
The concept of natural boundary conditions is also important in a more side is equal to - f F; '5u; dv. On account of the symmetry of <1;;, the second
gene: al type of vari ational problem in which boundary values occur explicitly integral may be written as
in the functionals. It can be generalized also to functionals involving severa!
dependent and independent variables and higher derivatives of the dependent fv<f;;[t(t5u;,; + '5u 1,;)] dv = fv°·;; '5e;; dv.
variables.
The idea of deriving natural boundary conditions for a physical problem Therefore, (1) becomes
is of great importance and will be illustrated again and again below. See,
for example, Sec. 10.8, 11.2. (2) ... ( F; '5u; dv + ( f: '5u; dS = ( <1;; '5e;; dv.
Jv Js" Jv
10. 7. THEOREM OF MINIMUM POTENTIAL This equation expresses theprinciple ofvirtual work. The surface integral needs
ENERGY UNDER SMALL VARIATIONS
only be integrated over S ,,, since '5u; vanishes over the surface S,. where
boundary displacements are given.
(v;) OF DISPLACEMENTS
lf the strain-energy function W (e11 , e12, ••• ) exists, so that <1;; = oW/oe;;,
(Sec. 6.1), then it can be introduced into the right-hand side of Eq. (2).
Let a body be in static equilibrium under
Since
the action of specified body and surface forces.
(Fig. 10.7:1). The boundary surface S shall be (3) r
Jv
<1;; '5e,; dv = raw '5e,; dv = t5 w dv,
Jva~; Jv
r
assumed to consist of two parts, S,, and S,,,
with the following boundary conditions. the principie of virtual work can be stated as
V
Over S,, : The surface traction T; is prescribed.
Fig. 10.7:1. Notations. (4) .&. t5 ( W dv - ( F; '5u; dv - ( f: '5u; dS = O.
Over S,,: The displacement u; is prescribed. Jv Jv Js,,
We assume that there exists a system of displacements u1 , u2 , u3 satisfying Further simplification is possible if the body force F, and the surface
the Navier's equations of equilibrium and the given boundary conditions. •
Let us considera class of arbitrary displacements u, + '5u, consistent with the
tractions T¡ are conservative so that
constraints imposed on the body. Thus i5u; must vanish over Su, but it is (5)
oG
F •. = - :::l-
•• = _ a
T, J
,
arbitrary over S,,. We further restrict i5u; to be triply differentiable and to be UU; ou, .
 286 ENERGY THEOREMS Chap. 10 Sec. 10.7 THEOREM OF MINIMUM POTENTIAL ENERGY 287

The functions G(u 1 , u 2 , u3) and g(ui, u 2 , u3) are called the potential of Fi -áud The sum of the terms in the first line on the right-hand side vanishes on
V
account of (4) . The sum in the second line is positive for sufliciently small
T¡*, respectively. In this case,
values of strain óei1 and can be seen as follows. Let us set ei; = O in Eq. (10).
The constant term is immaterial. The linear term must vanish because
(6) - { Fi óu; dv - { T¡ óui dS = ó { G dv + ó { g dS.
Jv Js 11 Jv Js oW/oei; = ai;, which must vanish as e;;-+ O. Hence, up to the second order,
Then Eq. (4) may be written as 1 o2 W
(12) - óei; óek 1 = W(óeii).
(7) .Á. ó"Y = O, 2 oe;; oekl
•.
where Therefore, (11) becomes

(8) ... "Y L

== (W + G) dv + f l dS (13) "Y' - "Y= f W(óe;;) dv.

The function "Y is called the potential energy of the system. This equation
states that the potential energy has a stationary value in a class of admissible
variations óui of the displacements u; in the equilibrium state. Formulated in
another way, it states that, of ali displacements satisfying the given, boundary
conditions, those which satisfy the equations of equilibrium are distinguished by
a stationary (extreme) value of the potential energy. For a rigid-body, W
vanishes and the familiar form is recognized. We emphasize that the linearity
of the stress-strain relationship has not been invoked in the above derivation,
so that this principie is valid for nonlinear, as well as linear, stress-strain law,
as long as the body remains elastic.
That this stationary value is a miminum in the neighborhood of the
natural, unstrained state follows the assumption that the strain energy
function is positive definite in such a neighborhood (see Secs. 12.4, 12.5). This
can be shown by comparing the potential energy "Y of the actual displace-
ments u; with the energy "Y' of another system of displacements u; óui
satisfying the condition óui = O over Su- We have
(9) "Y' - "Y = f [W(e 11 + óe11 , . • . ,) - W(e 11 ,
If W(óe; 1) is positive definite, then the last line in (11) is positive, and
(14) "Y' - "Y > O
and that "Y is a mínimum is proved. Accordingly, our principie is called the
principie of mínimum potential energy. The equality sign holds only if all
óe;; vanish, i.e., if the virtual displacements consist of a virtual rigid-body
motion. If there were three or more points of the body fixed in space, such
a rigid-body motion would be excluded, and "Y is a strong mínimum;
otherwise it is a weak mínimum.
To recapitulate, we remark again that the variational principie (2) is
generally valid; ( 4) is established whenever the strain energy function
W(e 11 , e12 , •• • ,) exists; and (7) is established when the potential energy "Y
can be meaningfully defined, but the fact that "Y is a minimum for "actual"
displacements is established only in the neighborhood of the stable natural
+ state, where W is positive definite.
Conversely, we may show that the variational principie gives the equations
of elasticity. In fact, starting from Eq. (7) and varying u;, we have

-J
. . • ,)] dv

Fi óui dv - f T¡ óui dS.

(15) ó"Y =
i-
V
oW Óe;; dv -
oeii
i V
,l
F; óui dv -
S
'* óu¡. dS
Ti = O.
But
Expanding W(e 11 + óe11 , ••• , ) into a power series, we have
oW
f uei
':l
óe" . dv = l2 fa ..(óu t,. '. + óu ,,i. .) dv = -fai..,, 1 óu . dv +fa.11·Y; óu . dS
i, i •

+ -oW óei;
1,

(10) W(e 11 + Óe11 , • •. ,) = W(e 11 , . . . ,)

1

oeii Hence,
+ - aw
2
1
Óe;; óek 1 + .... (16) fvcaii, 1 + F;) óui dv + fsca;1Y1 - T¡) ÓU; dS = o,
2 oeii oekl
A substitution into (9) yields, up to the second order in óei;, which can be satisfied for arbitrary óu; if

(11) "Y' - "Y =

i aw
V
-
Oe;;
óei; dv - J V
Fi óui dv - JtrS T.i* óui dS (17) ªi1 ,1 + Fi = O in V
and either
+ 1- a w
2

i V 2 Oei; Oekl
óei 1 óek 1 dv. (18) óui = O on S,.. (rigid boundary condition),
 288 ENERGY THEOREMS Chap. 10 Sec. 10.8 EXAMPLE OF APPLICATION: STATIC LOADING ON A BEAM 289

or
·~ 10.8. EXAMPLE OF APPLICATION: STATIC LOADING
(19) •• =
Ti Gi ;V; on S" (natural boundary condition). ON A BEAM-NATURAL AND RIGID END
The one-to-one correspondence between the differential equ'ations of CONDITIONS
equilibrium and the variational equation is thus demonstrated; for we have
first derived (7) from the equation of equilibrium and then have shown that, As an illustration of the application of the minimum potential energy
conversely, Eqs. (17)-(19) necessarily follow (7). principie in formulating approximate theories of elasticity, let us consider the
The commonly encountered externa! force systems in elasticity are approximate theory of bending of a slender beam under static loading. Let
V
the beam be perfectly straight and lying along the x-axis before the application
conservative systems in which the body force F; and surface tractions T;
of externa! loading, which consists of a distributed lateral load p(x) per unit
are independent of the elastic deformation of the body. In this case, "Y is
length, a bending moment M 0 , anda shearing force Q 0 at the end x = O, and
more commonly written as
a moment M 1 and a shear Q 1 at the end x = l (Fig. 10.8:1). We assume
that the principal axes of inertia of every cross section of the beam lie in two
(20) ,,y == r w dv - Jvr
Jv
F;U; dv - r f:u; dS.
Js" mutually orthogonal principal planes and that the loading p, M, Q are

A branch of mechanics in which the externa! forces are in general non-

dA
conservative is the theory of aeroelasticity. In aeroelasticity one is con-
cerned with the interaction of aerodynamic forces and elastic deformation.
The aerodynamic forces depend on the flow and the deformation of the entire
body, not just the local deformation; thus in general it cannot be derived from
a potential function. The principie of virtual work, in the form of Eq. (4), is
Mo~
c1 ~M1x~ ,p
still applicable to aeroelasticity.
ºº 01

Problem 10.1. The mm1mum potential energy principie states that elastic
equilibrium is equivalent to the condition J = min, where

(a) J = r[W(e ;;) + F;u ;] dV + Jsar

Jv
T;U; dS, (varying u;);
Fig. 10.8:1. Applications to a simple beam.

applied in one of the principal planes. In accordance with the approximate

(b) e;; = i(u;,j + ui,;). beam theory, we assume that every plane cross section of the beam remains
By the method of Lagrange multipliers, the subsidiary condition (b) can be in- plane during bending.
corporated into the functional J, and we are led to consider the variational equation Let ds be the are length along the neutral axis. Under these assumptions,
we see that when the neutral axis of the beam is bent from the initial straight
(c) {JJ' = 0, (varying e;;, u; independently),
line into a curve with radius of curvature R, the length of a filament,
where initially ds and parallel to the neutral axis, is altered by the bending in the ratio
(d) J' =J + r A;;[e;; - i(u;,; + U;,;)] dV.
+
1 : (1 r¡/ R), where r¡ is the distance between the filament and the neutral axis.
Jv The strain is r¡/R, and the force acting on the filament is E r¡dA/R, where dA
is the cross section of the filament. The resultant moment of these forces about
Since the quantity in [ ] is symmetric in i,j, we may restrict the Lagrange multi-
the neutral axis is
pliers A;; to be symmetric, A; ; = A;;, so that only six independent multipliers are
needed. Derive the Euler equations for (c) and show that A;; should be interpreted
as the stress tensor. JA
r¡ · E -r¡ dA
R
= -E
R
i
A
r¡ 2 dA = -El ,
R
Note: Once Lagrange's multipliers are employed, the phrase "minimum con-
ditions" used in the principie of minimum potential energy has to be replaced by where I is the moment of inertia of the area of the beam cross section. The
"stationary conditions." angle through which the cross sections rotate relative to each other is
290 ENERGY THEOREMS Chap. 10 Sec. 10.8 EXAMPLE OF APPLICATION : STATIC LOADING ON A BEAM 291
,.
ds/R. Hence, the work to bend a segment ds of the beam to a curvatur;; Since by is arbitrary in the interval (O,/), we obtain the differential equation of
1/R is the beam
l
2( 2
y)
El ds (5) -d 2 E dl -2 -p = O, Ü < X < /.
2 R2 ' dx dx
where the factor t is added since the mean work is half the value of the In order that the remaining terms in (4) may vanish, it is sufficient to have the
product of the final moment and angle of rotation. Hence, integrating end conditions
throughou~ the beam, we have the strain energy 2
(6a) Either El(dd Xy) - M ¡ = O or b( dy) = O.
(1) •. {1El ds. U = ! 2
l dx 1
2 R2 Jo 2

If we now assume that the deflection of the beam is infinitesimal so that (6b) Either El ( d y) - M O = O or b( dy) = O.
dx 2 o dx o
if y denotes the lateral deflection, the curvature 1/R is approximated by
2
d2y/dx2, and ds is approximated by dx, then
(6c) Either ~(El d y) - Q1 = O or by 1 = O.
51El (d2y)2 dx dx 2 1
(la) U = -1 - dx.
2 o
The potential energy of the externa! loading is, with the sign convention
dx 2
(6d) Either ~( El d2y)
dx dx 2 o
- ºº or by0 = O.
specified in Fig. 10.8: 1 and with y 0 , Yi, (dy/dx) 0 ,(dy/dx) 1 denoting the value
of y and dy/dx at x = O, !, respectively, If the deflection y 0 is prescribed at the end x = O, then by0 = O. If the slope
(dy/dx) 0 is prescribed at the end x = O, then b(dy/dx) 0 = O. These are called
(2) -Jp(x)y(x) dx + Mo(dy) - M 1(dy) - Q0 y 0 + Q y¡. "rigid" boundary conditions. On the other hand, if the value of y 0 is un-
1
dx o dx i specified and perfectly free, then by0 is arbitrary and we must have
Hence, the total potential energy is 2
y)
(3) "Y= -151El (d-dxy) dx - 5¡ py dx + M (dy)
2 o
2

2
2

dx o
0 -
o
(7) -d ( Edl -2 - Q0 =0
dx dx o
asan end condition for a free end; otherwise b"Y cannot vanish for arbitrary
- M1(~~) 1
- QoYo + QiYi- variations by0 • Equation (7) is called a "natural" boundary condition.
Similarly, all the left-hand equations in (6a)- (6d) are natural boundary
At equilibrium, the variation of "Y with respect to the virtual displacement by conditions, and ali the right-hand equations are rigid boundary conditions.
must vanish. Hence,

5¡ d 5¡ The distinction between natural and rigid boundary conditions assumes

b"Y =
2
(d
El~ b ~ dx -
2
)
p by dx
great importance in the application of the direct methods of solution of
o dx dx o variational problems; the assumed functions in the direct methods must
satisfy the rigid boundary conditions. See Sec. 11.8.
+M
b(dy) - M 1 b(dy) - Q0 by 0 + Q1 by 1 = O.
0 lt is worthwhile to consider the following question. The reader must be
dx o dx i
familiar with the fact that in the engineering beam theory, the end conditions
Integrating the first term by parts twice and collecting terms, we obtain
often considered are:
(4)
o dx 5
2
b"Y = ¡ [ -d 2( El -d2 2 - p by dx
dx
y) J (8a) Clamped end: y =0, dy = o,
dx
2 2
+ [E1(d y)2 - 1]
dx i dx I
[E1(d y)
dx 2 o
M b(dy) -
- 0] b(dy)
dx o
M (8b) Free end: El d y
2
= O ~(El d 2
2
y) = O,
2
dx ' dx dx
- [~dx E/(ddx 2
~) -
i
Q1] by 1 + [~
dx
El(ddx
2
~)
o
- Q0] dyo
(8c) Simply supported end: y = O, El d y
2
= O
=Ü dx 2 '
292 ENERGY THEOREMS Chap. 10 Sec. 10.9 THE COMPLEMENTARY ENERGY THEOREM 293

where y(x) is the deflection function of the beam. May we ask why a.rct the minimum principle. We pose the problem as in Sec. 10.7 with a body held in
other two combinations, namely - · ·� equilibrium under the body force per unit volume Fi and surface tractions per
(_9a) y=O, !!_(
dx
d
EI y2 ) = 0,
dx
1
unit area T,• over the boundary Sa, whereas over the boundary s.. the dis-
placements are prescribed. Let a,, be the "actual" stress field which satisfies
dy ds
EI y2 =0'
the equations of equilibrium and boundary conditions
(9b) =0'
dx dx in V,
never considered? (1)
An ace;eptable answer is perhaps that the boundary conditions (9a) and ai,,,, = Ti
y.
on Sa.
(9b) cannot•be realized easily in the laboratory. But a more satisfying answer Let us now consider a system of variations of stresses which also satisfy the
is that they are not proper sets of boundary conditions. If the conditions (9a) equations of equilibrium and the stress boundary conditions
or (9b) were imposed, then, according to (4), it can not at all be assured that
the equation lJ"i', = 0 will be satisfied. Thus, a basic physical law might be in V,
violated. These boundary conditions are, therefore, inadmissible.
(2) (()<1'1)11I = () r. on Sa,
From the point of view of the differential Eq. (5), one may feel that the end
conditions (9a) or (9b) are legitimate. Nevertheless, they are ruled out by the 15a,, are arbitrary on s...
minimum potential energy principle on physical grounds. In fact, in the In contrast to the previous sections, we shall now consider the complementary
theory of differential equations the Eq. (5) and the end conditions (8) are virtual work,
known to form a so-called self-adjoint differential system, whereas (5) and (9)
would form a nonself-adjoint differential system. Very great difference in
mathematical character exists between these two catagories. For example, a
f/' lJF, dv +Lu, 1'Yt dS,
free vibration problem of a nonself-adjoint system may not have an eigen­ which, by virtue of (2) and through integration by parts,
vector, or it may have complex eigenvalues or complex eigenvectors.
There are other conceivable admissible boundary conditions, such as to =- f/•{15aH),, dv +Luit5a")", dS
Iv Is
require
dy 2
= (15a")u,,, dv -Lui11,{tia") dS + uJ1'<1")11, dS
(10) = 0, !!_(EI d y) = 0, at X = 0.
dx dx dx2
Such an end, with zero slope and zero shear, cannot be easily established in
the laboratory. Similarly, it is conceivable that one may require that at the
= !fv(1'<1")(u,., + u1,J dv
end x = 0 the following ratios hold:
=J/H ()<1Hdv .
lJ (ay).. 15y = c, a constant,

f/" <5aH dv = f/' <5F, dv + f/' <57; dS.

ax Hence,

[!
(11)
(EI��) - Q0]: [EI�� - M0] = c, the same constant. ( 3) A

This pair of conditions are also admissible, but are unlikely to be encountered
in practice.
10.9. THE COMPLEMENTARY ENERGY THEOREM
UNDER SMALL VARIATIONS OF STRESSES
In contrast to the previous sections let us now consider the variation
of stresses in order to investigate whether the "actual" stresses satisfy a
This equation may be called the principle of virtual complementary work.
Now, if we introduce the complementary strain energy Wc,f which is a
function of the stress components <111 , <111, • • • , and which has the property
that,
(4)
t The Gibbs' thermodynamic potential (Sec. 12.3) per unit volume, p(!), is equal to
the negative of the complementary strain energy function. If the stress-strain law were
linear, then W0(1111) and W(e11 ) are equal: -p(!) = W0 = W (linear stress-strain law).
 294 ENERGY THEOREMS Chap. 10 Sec: 10.9 THE COMPLEMENTARY ENERGY THEOREM 295

then the complementary virtual work may be written as Before we proceed further, it may be well worthwhile to consider the
. .,, concept of complementary work and complementary strain energy. Consider
(5) J V
ui oFi dv + f
S
ui oT; dS = J
V
awc OCli; dv = o { Wc dv . .
ªCli; Jv
a simple, perfectly elastic bar subjected to a tensile load. Let the relationship
between the load P and the elongation of the bar u be given by a unique curve
as shown in Fig. 10.9: l. Then the work W is the area between the displace-
Since the volume is fixed and ui are not varied, the result above can be ment axis and the curve, while the complementary work Wc is that included
written as between the force axis and the curve. Thus, the two areas complement each
other in the rectangular area (force)· (displacement), which would be the
(6) ... o"/'* = O,
work if the force were acting with its full intensity from the beginning of the
~ .
where "/'*, as a function of the stresses C111 , C112 , ••• , the surface traction Ti
and the body force per unit volume Fi> is defined as the complementary energy

(7) ... "f"*(C111, •.• , Fi) == fvwc dv - f/iFi dv - Is uiT; dS.

In practice, we would like to compare stress fields which all satisfy the
p <f
O'&
equations of equilibrium, but not necessarily the conditions of compatibility.
V

In other words, we would have oFi = O in V and oTi = O on S,,. In this case
V

OCli; and, hence, oTi are arbitrary only on that portion of the boundary where
displacements are prescribed, S.,. Therefore, only a surface integral over u ~
-j ¡.-<fr
S., is Ieft in the left-hand side of (5) and we have (o) (b)

(8) ... "f"*(C111 , ••• , C133 ) == { Wc dv

Jv
-f s ..
u}; dS.
Fig. 10.9:1. Complementary work and strain energy.

displacement. Naturally, W and Wc are equal if the material follows Hooke's

Therefore, we have the law.
The principie of mínimum potential energy was formulated by Willard
THEOREM. Of ali stress tensor fields Cli; that satisfy the equation of equilib- Gibbs; and many beautiful applications were shown by Lord Rayleigh. The
rium and boundary cond~tions where stresses are prescribed, the "actual" one complementary energy concept was introduced by F. Z. Engesser; its appli-
is distinguished by a stationary (extreme) value of the complementary energy cations were developed by H. M. Westergaard. In the hands ofKirchhoff, the
"/'( C111, ••• , C133) as given by (8). mínimum potential energy theorem becomes the foundatjon of the approxi-
mate theories of plates and shells. Recently, Argyris has made the comple-
In this formulation, the linearity of the stress-strain relationship is not mentary energy theorem the starting point for practica! methods of analysis
required, only the existence of the complementary strain energy function is of complex elastic structures using modern digital computers. See Bibliog-
assumed. However, if the stress-strain law were linear, and the material is raphy 10.2, p. 494.
isotropic and obeying Hooke's law, then Let us now return to the complementary energy theorem. We shall show
that in the neighborhood of the natural state, the extreme value of the comple-
1+V mentary energy "/' is actually a minimum. By a natural state is meant a state
(9) Wc(C111, · · ·, Claa)
V
= - - (Claa)
2E
2
+- - ClüCl;;.
2E of stable thermodynamic existence (see Sec. 12.4). In the neighborhood of a
natural state, the thermodynamic potential per unit volume p<l> can be
We must remark that the variational Eqs. (10.7 :2) and Eq. (3) of the approximated by a homogeneous quadratic form of the stresses. For an
present section are applicable even if the body is not elastic, for which the isotropic material, - p<l> is given by Eq. (9), in which Wc is positive
energy functional cannot be defined. These variational equations are used in defi.nite.
the analysis of inelastic bodies. The proof that "/'* is a mínimum is analogous to that in the previous
 296 ENERGY THEOREMS Chap. 10 Sec. 10.9 THE COMPLEMENTARY ENERGY THEOREM 297

section, and it can be sketched as follo ws. Comparing f *(<1i; + b<1i1) with The variations b<1i; are subjected to the restrictions
f *(<1i;), we have .~
( 14) (b<1i;),; = o in V,
(10) f * ( <111 + b<111, · · • ,) - f *(<111, • • • ,)
(15) (b<l;;)'V; = o on S,,.

= Ív [W.( <111 + b<111, .. . ,) - Wc( <111 , . .. ,)Jdv To accommodate the restrictions (14) into ( 13), we make use of the celebrated
result that the equations of equilibrium ( 14) are satisfied formally, as can be
easily verified, by taking
1
V V

- [~( <111 + b<111, . . . ,) - ~( <111, . • . ,) ]ui dS

Su
•. b<111 = + </>33 ,22 -
</>22,33 2</>23,23•

= Ív [(1 ! v <1ii - i fJ bii) b<1i; + W.( b<111 , ... ,)Jdv b<122 = <p33 ,11 + </>11 ,33 -
b<133 = </>11 ,22 + </>22,11 -
2</>31,31,

2</>12,12 ,
(16)
b<123 = <p31 ,12 + </>12 ,13 - </>11 ,23 - </>23,11•
- ( (b<1i;)v 1ui dS
Jsu b<131 = </>12 ,23 + </>23,21 - </>22,31 - <p31 ,22,

b<112 = </>23 ,31 + <p31 ,32 -

= Í/ii b<1ii dv -Isu(b<1i1)'V;Ui dS +JW.(b<111, .. . ,) dv <p33 ,12 - </>12,33•

where <p;; = </>;i are arbitrary stress functions. On setting <p12 = <p 23 = <p 31 =

= bf * + L W.(b<1 11 , b<112 , . . . ,) dv,

O, we obtain the solutions proposed by James Clerk Maxwell. On taking
</> 11 = <p 22 = <p 33 = O, we obtain the solutions proposed by G . Morera.
Let us use the Maxwell system of arbitrary stress functions for the varia-
where
tions b<1;;· Equation (13) may be written as
1 + 'JI 'JI 2
(11) Wc(b<111, ... ,) = - - (b<1i;)(b<1i;) - - (b<1i;) > O.
2E 2E

Wo(b<1 11 , ... ,) is positive definite for infinitesimal variations b<1w Hence,

(17) bf * = L + <p3s,22) + e2i <p33,11 + </>11,33)
[e11( </>22,33

when bf * = O, + e3a( </>11,22 + </>22,11) - 2e2s</>11, 23 - 2e31 </> 22, 31 - 2e12 <p33,12J dv

(12) f*(<1 11 + b<111 , .. . ,) - f*(<1 11 , ••• ,) > O, - ( ui bT; dS

Jsu
and that "Y*( <111 , • • • ,) is á mínimum is proved. = Ü.
The converse theorem reads:
Integrating by parts twice, we obtain
Let f* be the complementary energy de.fined by Eq. (8). 1f the stress
tensor .field <1i; is such that bf* = Ofor ali variations of stresses b<1i; which (18) bf* = fv[(e22.s3 + e33,22 - 2e23,23)<p11 + (e33,11 + eu,33 - 2e31,31)</>22
satis/y the equations of equilibrium in the body and on the boundary where
surface tractions are prescribed, then <1H a/so satis.fies the equations of com- + (en ,22 + e 22, 11 - 2e12 ,12)<p 33] dv + a surface integral

patibility. In other words, the conditions of compatibility are the Euler = Ü.

equation for the variational equation bf* = O.
Inasmuch as the stress functions <p11 , <p 22 , <{> 33 are arbitrary in the volume V,
the Euler's equations are
The proof was given by Richard V. Southwell1.2 (1936) through the appli-
cation of Maxwell and Morera stress functions. We begin with the variational
equation
(19) e22 ,33 + e33,22 - 2e23, 23 = O,
etc., which are Saint-Venant's compatibility equations (see Sec. 4.6). The
(13) bf* = ( e;; b<1i 1 dv - ( ui b'I'¡ dS = O.
Jv Jsu treatment of the surface integral is cumbersome, but it says only that over
298 ENERGY THEOREMS Chap. 10 Sec. 10.10 REJSSNER'S PRINCIPLE 299

S,.. the values of ui are prescribed and the stresses are arbitrary; the derivation The six equations (24), (25), i = 1, 2, 3, require six Lagrange multipliers A;,µ;
concerns a certain relationship between </>;; and their derivatives -Á:u be which are functions of (x1 , x 2 , x 3). Show that the Euler equations for (27) yield the
satisfied on the boundary. physical interpretation
Similarly, the use of Morera system of arbitrary stress fünctibns leads to (28) A; = U;, µi = U;.
the other set of Saint-Venant's com patibility equations
(20) eu,2a = - e2a,11 + ea1,21 + e12,a1, 10.10. REISSNER'S PRINCIPLE
etc. [Eq. (4.6 :4)).
lf we ,start with Ir//* in the form Consider the functional J(ei; , uk, ai;), where ai; = a;;:

(21) ó"// * = O = f (1 +
V
- E')J a;; - -E
')J ªªªÓ;; ) Óa;; dv + J
Su
u; óT;
V
dS
(1) J = fv[W(e;;) - F;ui] dv - L a;;[ei; - t(u;, ; + u;;)] d1,

and introduce the stress functions, the Beltrami-Michell compatibility

equations
- Jsr T/ui dS
11
-J Su
ai;V;(ui - u7) dS, (S = S11 + S,,).
Let us seek the necessary conditions for J to be stationary. On setting the
(22) V2a;,· + -1 +1- v 8,;,· + _v_
l
ó;,.Fk + F; ,· + F,. ; = O
- ')J •
k
• •
in V
first variation of J to zero, permitting the fifteen functions e;;, ui, and ai;,
(i = 1, 2, 3) to vary over the domain V, and u; to vary over S 11 , anda;; to vary
can be obtained directly. V

In concluding this section, we remark once more that when we consider over S,,, while F;, T;*, and u¡ are prescribed, we obtain
the variation of the stress field of a body in equilibrium, the principie of
virtual complementary work has broad applicability. The introduction of the
complementary energy functional "//*, however, limits the principie to
(2) O= óJ =l V ue;;
{ºW
:l "'' - F.1. óu t. -
óe. óu.1.,.. [ei,.. - 1
2 (u t,. ; + u ,.,
. .)]

elastic bodies. That "//* actually is a mínimum with respect to all admissible - a; 1 [óe;; - t(óu;,; + Óu;.i)]} dv
variations of stress field is established only if the complementary strain
energy function is positive definite.
There are many fascinating applications of the mínimum complementary
-
J V*
S 11
T; Óu; dS - J Su
[óai;v ;(ui - ui* ) + ai;v; Óu;] dS.
energy principie. In Sec. 10.11 below we shall consider its application in Integrating by parts those terms involving Óu;,;, we obtain the Euler equations:
proving Saint-Venant's principie in Zanaboni's formulation.
(3) oW = a .. in V
'.l
ue;; " '

Problem 10.2. The principie of virtual complementary work states that

(4) a;;,;+ F; = O in V,

(23) r
Jv
e ;;C)a ;; dv - r uli5T; dS = o
Jsu
(5) e;; = t(u;,; + u;,;) in V,
V

under the restrictions that (6) a;;'V ; = T;* on S11 ,

(24) óaii,; = O in V, (7) U; = U; ,* Óu; = O on S,...

(25) Óa;;V; = Ü 00 Sa. Clearly these are the basic equations of linear elasticity (Section 7.1 ).
The Eq. (2), óJ = O, where J is defined in (l), is a statement of Reissner's
(26) u; = uf prescribed on S,,, but óT; = óa;;v; arbitrary on S,,.
principie as applied to linear elasticity.
Using Lagrange multipliers, we may restate this principie as The motivation for J is clear. The functional

(27) r
Jv
e;;Óa;; dv - r u¡r5T;dS + Jvr
Jsu
A;Óa;;,; dv - r
Jsa
µ;Óa;;V; dS = o 1 1 = { {W(e;;) - Fiu;} dv
Jv
-J T;*u; dS
S 11
300 ENERGY THEOREMS Chap. 10 Sec. 10.11 SAINT-VENANT'S PRINCIPLE 301

represents the sum of the strain energy and the potential energy o[ the
externa! loads. Now if we wish to find a stationary value of J 1 by varyir1g e;;
and u;, subjected to the subsidiary conditions (5), (6), and (7), we may do so
by introducing Lagrange multipliers. Obviously a ;; plays the role of the
Lagrangian multiplier in (1).
Eric Reissner. 10 · 2 announced this principie in 1950 and extended it to
finite elastic deformations in 1953. In Reissner's original paper, the com-
plementary energy functional was used instead ofthe potential energy. In the
infinitesimal displacement theory, the complementary energy is defined as
•.
(8) Wc( ai;) = ai;e;; - W(e;1).
Wc( O";;) is a function of the stress components if e;; on the right-hand side are
expressed in terms of ak 1by means of Hooke's law. W 0 (a;;) has the property
that
which the couple is the resultant. Such a principle is nearly always applied,
consciously or unconsciously, when we try to simplify or idealize a problem in
mathematical physics. It is used , for example, in devising a simple tension
test for a material, when we clamp the ends of a test specimen in the jaws of a
testing machine and assume that the action on the central part of the bar is
nearly the same as if the forces were uniformly applied at the ends.
The justification of the principie is largely empirical and, as such, its
interpretation is not entirely clear.
One possible way to formulate Saint-Venant's principie with mathematical
precision is to state the principie in certain sense of average as follows
(Zanaboni, 1 º· 3 1937, Locatelli, 10 ·3 1940, 1941). Consider a body as shown
in Fig. 10.11:1. A system of forces Pin static equilibrium (with zero
resultant force and zero resultant couple) is applied to a region of the body
enclosed in a small sphere B. Otherwise the body is free. Let S' and S" be

(9) aw. = e;;• s' ,J:

00";;
~)a
On substituting W = O";;e;; - W 0 from Eq. (8) into Eq. (1), we have Reissner's R'
functional J R: R'
s" ~f (' ~

(10) JR = L
{-W0 (a;;) - F;u; + fa; 1(u;,; + u;,;)} dv ~}8 C§2JfB
r T;*u ; dS -J O";;V;(U; -
R"
- Jsu Su
u:) dS Fig. 10.11 :1 Fig. 10.11 :2

and Reissner's principie that the elastic equilibrium is distinguished by
CJJR = O, when O"; ; and u; (i,j = 1, 2, 3) are varied independently, provided
that the tensor a;, is symmetric.
lf we vary JR with respect to O";; only, (c5u; =
O), keeping the stress
conditions (4) and (6) satisfied, then the complementary energy theorem is
obtained.
10.11. SAINT-VENANT'S PRINCIPLE
In 1855, Barre de Saint-Venant enunciated the "principie of the elastic
equivalence of statically equipollent systems of loads." According to this
principie, the strains that are produced in a body by the application, to a
two arbitrary nonintersecting cross sections, both outside of B, with S" farther
away from B than S'. Dueto the system ofloads P, stresses are induced in the
body. lf we know these stresses, we can calcula te the tractions acting on the
surfaces S' and S". Let the body be considered as severed into two parts at S',
and let the system of surface tractions acting on the surface S' be denoted by
R' which is surely a system of forces in equilibrium (see Fig. 10.11 :2). Then a
convenient measure of the magnitude of the tractions R' is the total strain
energy that would be induced in the two parts should they be loaded by R' alone.
Let this strain energy be denoted by UR'· We have,
U R' = L
W(a!fl) dv,

small part of its surface, of a system of forces statically equivalent to zero
force and zero couple, are of negligible magnitude at distances which are
large compared with the linear dimensions ofthe part. When this principie is
applied to the problem oftorsion of a long shaft dueto a couple applied at its
ends, it states that the shear stress distribution in the shaft at a distance from
the ends large compared with the cross-sectional dimension of the shaft will
be practically- independent of the exact distribution of the surface tractions of
where W is the strain energy density function, and the stresses a\:l correspond
to the loading system R'. Similarly, Jet the magnitude of the stresses at the
section S" be measured by the strain energy UR'• which would have been
induced by the tractions R" acting on the surface S " over the two parts of
the body severed at S ". Both UR , and UR. are positive quantities and they
vanish only if R', R" vanish identically. Now we shall formulate Saint-
Venant's principie in the following form (Zanaboni, 1 º· 3 1937).
 Seé. 10.11 SAINT-VENANT' S PRINCIPLE 303
302 ENERGY THEOREMS Chap. 10
Now tbe system of forces R represents the interna) normal and shear stresses
Let S' and S" be two nonintersecting sections both outside a sphere JJ. !f
acting on the interface S of the body C1 + C 2 • lt is therefore determined by
the section S " líes at a greater distance than the section S' from the spheré .B, in
the mínimum complementary energy theorem. Considera special variation of
which a system of self-equilibrating forces P acts on the body, the.n .
stresses in w hich all the actual forces R are varied in the ratio 1 : ( 1 + e),
(1) UR. < UR'. wbere e may be positive or negative. Tbe work UJu will be changed to
(1 + e) 2 Uj 1 , because the load and deformation wilt both be changed by a
In this form, the diminishing influence of the self-equilibrating system of factor (l + e). Similarly, Uj 2 is changed to (l + 1:) 2 u; 2. But UJ;R is only
loading P as the distance from B increases is expressed by the functional UR• changed to (1 + e)UJ;R, because the load Pis fixed, while the deformation
wbich is.ª special measure of tbe stresses induced at any section outsidé B. is varied by a factor (1 + e). Hence, U1+2 is changed to
The reasoll for the choice of UR as a measure is its positive definiteness
character and the simplicity with wbich the theorem can be proved. Furtber (4) U 1,+ 2 = U1 + (1 + e)2 URl* + (1 + e)2UR2
* + (1 + e)UPR·
*
sharpening of tbe principie will be discussed later. The difference between (4) and (3) is
In order to prove the theorem (1), we first derive an auxiliary principie.
Let a self-equilibrating system of forces P be applied to a limited region B at LiU1+ 2 = 1:(2u;1 + 2u;2 + u;R) + 1: 2cu;1 + u;2).
the surface of an otherwise free elastic body Ci, (Fig. 10.11 :3). Let U1 be the
strain energy produced by Pin C1 . Let us now consider an enlarged body For U 1+ 2 to be a mínimum, LiU1+ 2 must be positive regardless of the sign of
C1 + C 2 by affixing to C1 an additional body C2 across a surface S which does e. This is satisfied if
not intersect the region B. When Pis applied to the enlarged body C1 + C2 ,
(5) *
2UR1 + 2UR2* + U pR
* = O.

CE]
,, tbe strain energy induced is denoted by U1+ 2 • Then
the lemma states that On substituting (5) into (3), we obtain

52
E!5.-/e
51
1

,' (2) U1+ 2 < U 1.

(6) U1+2 = U1 - (U;1 + u;2),
Proof of the Lemma. To compute U1+2 , we Since Uj 1 and UJ'i 2 are positive definite, we see that lemma (2) is proved.
p
8 (I;f_
C1 ~ )
imagine that the stresses in the enlarged body
-- 8 C1 + C 2 are built up by the following steps (see
Fig. 10.11 :3). First the load P is applied to C1 •
Now we shall prove Saint-Venant's principle embodied in Eq. (1).
Consider an elastic body consisting of three parts C1 + C2 + C 3 loaded by
Pin B , as shown in Fig. 10.11: l. Let this body be regarded first as a result of
-.f
The face S1 of C1 is deformed. Next a system of adjoining C2 + C 3 to C1 with an interface force system R', and then as a

CI~ \R
;
:::. 1 surface tractions R is applied to C1 and C2 on
_/é the surfaces of separation, S1 and S 2 • R will be
so chosen that the deformed surfaces S 1 and S 2
result of adjoining C 3 to C1 + C 2 with an interface force R". We have, by
repeated use of (6),

fit each other exactly, so that displacements of U 1+(2+3) = + U ;'(2+3)),

U 1 - ( U ;.1
~ ') material points in C1 and C2 are continuous as
Uc1+2>+ a = U1+2 - (U ; "<1+2> + U; .. a)
~'8 welt as the stresses. Now C1 and C2 can be brought
5
together and welded, and S becomes merely an = U 1 - (U;1 + u;2) - (u; . c1+2> + u;. a) .
Fig. 10.11 :3 interface. The result is the same if C1 and C2 Equating these expressions, we obtain
were joined in tbe unloaded state and the com-
bined body C1 + C2 is loaded by P. *
U R'l + U R'(2+3)
* * + U R2
= U R1 * + U R"C1+2)
* + u*R"3•
Tbe strain energy U1+2 is the sum of the work done by the forces in the
above stages. In the first stage, the work done by P is U1 . In the second or, since UJ'i 1 and UJ'i2 are essentially positive quantities,
stage, the work done by R on C2 is u; 2 ; the work done by R on C1 consists of
two parts, U} 1 if C1 were free, and UJ;R, the work done by the system of (7) *
U R'l *
+ U R'(2+3) > *
U R"(1+2) + u*R"3·
loads P due to the deformation caused by R. Hence, Tbis is Eq. (1), on writing UR,for Uh + U}c 2+a>, etc. Hence, the principie is
* + UR2
* + UPR·
*· proved.
(3) U1+2 =U¡+ URl
304 ENERGY THEOREMS
10.12. SAINT-VENANT'S PRINCIPLE-BOUSSINESQ-
VON MISES-STERNBERG FORMULATION
The Saint-Venant principie, as enunciated in terms of the strain energy
functional, does not yield any detailed information about individual stress
components at any specific point in an elastic body. However, such in-
formation is clearly desired. To sharpen the principie, it may be stated as
follows (v~n Mises, 1945). "If the forces acting upon a body are restriced to
severa! small parts of the surface, each included in a sphere of radius E, then
the strains and stresses produced in the interior ofthe body ata finite distance
from ali those parts are smaller in order of magnitude when the forces for
each single part are in equilibrium than when they are not."
The classical demonstration of this principie is dueto Boussinesq (1885),
who considered an infinite body filling the half-space z > O and subjected to
severa! concentrated forces, each of magnitude F, normal to the boundary
z = O. lf these normal forces are applied to points in a small circle B with
diameter E, Boussinesq proved that the largest stress component at a point P
which lies at a distance R from B is
(1) of order F/R 2 if the resultant of the forces is of order F,
(2) of order ( E/ R)(F/ R 2) if the resultant of the forces is zero,
(3) of order ( E/ R) 2(F/ R 2) if both the resultant force and the resultant
moment vanish.
(Cf. Secs. 8.8, 8.10.)
These relative orders of magnitude were believed to have general validity
until von Mises (1945) showed that a modification is necessary.
Consider the half-space z > O again. Let forces of magnitude F tangent
to the boundary z = O be applied to points in a small circle B o[ diameter E.
Making use of the well-known Cerruti solution (Sec. 8.8), von Mises obtained
~ -~~r -~Qr -F~
-F
{-F
(a) (b) (el (d)
Fig. 10.12:1. von Mises' examples in which forces tangential to the surface
of an elastic half-space are applied in a small area.
the following results for the four cases illustrated in Fig. 10.12: l. The order
of magnitude of the largest stress component at a point P which lies at a
distance R from B is
(1) of order a0 = F/R 2 in case (a),
(2) of order ( E/R)a0 in case (b),
(3) of order ( E/ R)aOin case (c),
(4) of order ( E/ R) 2a 0 in case (d).
Chap. 10 Sec. 10.12 BOUSSINESQ· VON MISES- STERNB ERG FORMU LATION 305
\. The noteworthy case is (c), which is drastically different from what one
, would expect from an indiscriminating generalization of Boussinesq's
result, for in this case the forces are in static equilibrium, with zero moment
about any axes, but all one could expect is a stress magnitude of order no
greater than ( E/R)a0 , not ( E/R) 2a 0. von Mises found that in this case the
order of magnitud e of the largest stress component is reduced to ( E/ R) 2a 0 if
and only if the externa! forces acting u pon a small part of the surface are such
as to remain in equilibrium when ali the forces are turned through an arbitrary
angle. (Such a case is called astatic equilibrium.)
von Mises examined next the stresses in a finite circular disk dueto Joads
acting on the circumference, and a similar conclusion was reached. (See Fig.
10.12 :2.)
These examples show that Saint-Venant's principie, as stated in the
traditional form at the beginning of this section, does not hold true.
-F+j€f+
J. S 4.::1
F
la) (b) (el
_ _ _-2F
(d)
Fig. 10.12:2. von Mises' examples of a circular disk subjected to loads on
the circumference.
Accordingly, von Mises 10 ·3 proposed, and Jater Sternberg16 ·3 proved (1954),
the following mathematical statement of the Saint-Venant princi ple:
Let a body be acted on by surface tractions which are finite and are
distributed overa number of regions ali no greater than a sphere of diameter
E. Consider an interior point x whose distance to any of these loading areas
in no less than a characteristic length which will be taken as unity. E is
nondimensionalized with respect to this characteristic Jength. Then, as E - O,
the order of magnitude of the strain components at x is as follows:
e(x, E) = O( EP)
where
( a) If the tractions have nonvanishing vector sums in at least one area, then
 306 ENERGY THEOREMS
Chap. 10
in general, p > 2. (Note that the surface traction is assumed tQ be finite,
so the resultant force --+ O as E2, since the area on wliich ·the,surface
tractions act--+ O as E2.) • when yielding occurs. Although no precise proof is available, Goodier10·3
(b) If the resultant of the surface tractions in every loading area vanishes,
then p > 3.
(c) If, in addition, the resultant moment of the surface tractions in every
loading area also vanishes, then still we can be assured only of p ;;;;, 3.
(d) p ;;;;, 4 in case of astatic equilibrium in every loading area, which may be
-d,escribed by the 12 scalar conditions
f T/ dS = O, f T/x; dS = O, (i,j = 1, 2, 3)
Js(,) Js<,>
for each loading area S(E), where T;* represents the specified surface
traction over S( E).
If the tractions applied to S(E) are parallel to each other and not tan-
gential to the surface, then if they are in equilibrium they are also in astatic
equilibrium, and the condition p > 4 prevails.
If, instead of prescribing finite surface traction, we consider finite forces
being applied which remain finite as E--+ O, then the exponent p should be
replaced by O, 1, 2 in the cases named above, as was illustrated in Figs. 10.12:1
and 10.12:2.
The theorem enunciated above <loes not preclude the validity of a stronger
Saint-Venant principie for special classes of bodies, such as thin plates or
shells or long rods. With respect to perturbations that occur at the edges of a
thin plate or thin shell, a significant result was obtained by K.O. Friedrichs1º· 3
(1950) in the form of a so-called boundary-layer theory. With respect to
lateral loads on shells, Naghdi 1 º·3 (1960) obtained results similar to von Mises-
Sternberg's.
10.13. PRACTICAL APPLICATIONS OF SAINT-VENANT'S
PRINCIPLE
It is well-known that Saint-Venant's principie has its analogy in hydro-
dynamics and that these features are associated with the elliptic nature of the
partía! differential equations. Ifthe differential equations were hyperbolic and
two-dimensional, local disturbances may be propagated far along the char-
acteristics without attenuation. Then the concept of the Saint-Venant's
principie will not apply. For example, in the problem of the response of an
elastic half-space to a line load traveling at supersonic speeds over the free
surface, as discussed in Sec. 9.7, the governing equations (9.7:11) are hyper-
bolic, and we know that any fine structure ofthe surface pressure distribution
is propagated all the way to infinity.
On the other hand, one feels intuitively that the validity of Saint-Venant's
Sec. 10.13 PRACTICAL APPLICATIONS OF SAINT-VENANT'S PRINCIPLE 307
principie is not limited to linear elastic solid or infinitesimal displacements.
One expects it to apply in the case of rubber for finite strain or to steel even
(1937) has argued on the basis of energy as follows:
Let a solid body be loaded in a small area whose linear dimensions are of
order E, with tractions which combine to give zero resultant force and couple.
Such a system of tractions imparts energy to the solid through the relative
displacements of the points in the small loaded area, because no work is done
by the tractions in any translation or rotation of the area as a rigid body.
Let the tractions be of order p. Let the slope of the stress strain curve of the
material be of order E. (The stress-strain relationship does not have to be
linear.) Let one element of the loaded area be regarded as fixed in position
and orientation. Then since the strain must be of order p/E, the displacements
of points within the area are of order (pE/E). The work done by the traction
acting on an element dS is of order (pEJJ · dS/E). The order of magnitude of
total work is, therefore, p 2 E3 /E. Since a stress of order p implies a strain
energy of order p 2/E per unit volume, the region in which the stress is of order
p must have a volume comparable with E3 • Hence, the influence of tractions
cannot be appreciable at a distance from the loaded area which are large
compared with E.
Goodier's argument can be extended to bodies subjected to limited
plastic deformation. In fact, the argument provides an insight to practica!
judgement of how local self-equilibrating tractions should influence the
strain and stress in the interior of a body.
An engineer needs to know not only the order of magnitude comparison
such as stated in von Mises-Sternberg theorem; he needs to know also how
numerically trustworthy Saint-Venant's principie is to his particular problem.
Hoff 1 º· 3 (1945) has considered severa) interesting examples, two of which are
given in Figs. 10.13:1 and 10.13:2 and will be explained below.
In the first example, the torsion of beams with different cross sections is
considered. One end of the beam is clamped, where cross sectional warping
is prevented. The other end is free, where a torque is applied by means of
shear stresses distributed according to the requirements of the theory of pure
torsion. The difference between the prescribed end conditions at the clamped
end from those assumed in Saint-Venant's torsion theory (Sec. 7.5) may be
stated in terms of a system of self-equilibrating tractions that act at the
clamped end. Timoshenko has given approximate solutions to these problems.
Hoff's example refers to beams with dimensions as shown in the figure and
subjected to a torque of 10 in. lb. Due to the restrictions of warping,
normal stresses are introduced in the bar in addition to the shearing stresses of
Saint-Venant's torsion. For the rectangular beam, the maximum normal
stress at the fixed end is equal to 157 lb/sq in. For the other two thin-walled
channel sections, the maximum normal stresses at the fixed end are 1230 and
10,900 lb/sq in., respectively, for the thicker and thinner sections. In Fig.
308 ENERGY THEOREMS
Chap. 10 Sec. 10.13 PRACTICAL APPLICATIONS OF SAINT-VENANT'S PRINCIPLE 309

10.13: 1 curves are shown for the ratio of maximum normal stress finax (x) in
any section x divided by the maximum normal stress f maxCO) in the fixed-end
section, plotted against the ratio distance x ofthe section from the fixed end of
the bar, divided by the total length L of the bar. Inspection of the curves
reveals that, while in the case of the solid rectangular section the normal
stress caused by the restriction to warping at the fixed end is highly localized,
it has appreciable values over the entire length of the channel section bars.
Consequently, reliance on Saint-Venant's principie in the calculation of

3.. 1-º 0 .6
Crossb 2"
section
of bar
1

0.032
,,
T(
l
2

fl
0.114"
3 stresses caused by torsion is entirely justified with the bar of rectangular
section. In contrast, sttesses in the thin-walled section bars depend largely
upon the end conditions.
Hoff's second example refers to pin-jointed space frameworks. In one
case a statically determínate framework is considered, to one end of which a
o .. ~1''..j 1"..J H set of four self-equilibrating concentrated loads are applied, as shown in Fig .
.._E o
.._E 0.5" 10.13 :2. The figures written on the elements of the framework represent the
0.4 forces acting in the bars measured in the same units as the applied loads.
Negative sign indicates compression. It can be seen that the effect of the
forces at one end of the structure is still noticeable at the other end.
Hoff's examples show that Saint-Venant's principie works only ifthere is a
possibility for it to work; in other words, only if there exist paths for the
1 b , 1 , r 1 1 :::::r:::;:;;:, 1 interna) forces to follow in order to balance one another within a short
O 0.2 0.4 0 .6 0 .8 1.0 distance of the region at which a group of self-equilibrating externa! forces is
x/L applied. This point of view is in agreement with Goodier's reasoning.
Fig. 10.13:1. Hoff's illustration of St-Venant's principie. Hoff's examples are not in conflict with the von Mises-Sternberg theorem,
for the latter merely asserts a certain order of magnitude comparison for the
stress and strain as the size of the region of self-equilibrating loading shrinks
to zero, and does not state how the stresses are propagated. On the other
hand, although Goodier's reasoning does not provide a definitive theorem, it
is very suggestive in pointing out the basic reason for Saint-Venant's principie
and can be used in estimating the practica! efficiency of the principie.

PROBLEMS

10.3. Derive the differential equation and permissible boundary conditions for a
membrane stretched over a simply connected regular region A by minimizing the
functional / with respect to the displacement w; p(x, y) being a given function

I = i JL(w: + w:) dx dy - JLpw dx dy.

10.4. Obtain Euler's equation for the functional

"°º
Fig. 10.13 :2. Hoff's example illustrating the slow decay of self-equilibrating
I' = ! JLcw: + w: + ('v 2 w + p) 2pw] dx dy.
2
-

forces in a space framework. Ans. 'v 4 w - 'v 2 w + 'v2p - p = O.

 Chap. 10 PROBLEMS 311
310 ENERGY THEOREMS Chap. 10

Note: Compare I' with I of Prob. 10.3. Since \7 2 w +p

= O for a memb¡ ane, 10.8. Let D be the set of ali functions u(x) with the following properties:
the solution of Prob. 10.3 also satisfies the present problem. But the intesrand of,/' (1) u(x) = a 1 sin 1rx + a 2 sin 21rx,
contains higher derivatives of w and the equation M ' = O is a sharper condition
D: (2) a 1 , a 2 are real, arbitrary numbers,
than the equation M = O. These examples illustrate Courant's method Óf sharpen-
{
ing a variational problem and accelerating the convergence of an approximating (3) a! + a~ > O.
sequence in the direct method of solution of variational problems.
Consider the functional 1
10.5. Bateman's principie in fluid mechanics states that in a flow of an ideal, { [u'(x)] 2 dx
nonviscous fluid (compressible or incompressible) in the absence of external body
forces, the ·:_pressure integral"

-fff pdxdydz,
L
J[u] = --=-=1---
lu(x)]2 dx

for ali functions u(x) in D .

which is the integral of pressure over the entire fluid domain, is an extremum. Questions:
Consider the special case of a steady, irrotational flow of an incompressible fluid, (a) Give necessary conditions for a function u*(x) in D so that u*(x) furnishes an
for which the Bernoulli's equation gives extremum of the functional J.
pV2
p = const. - 2 = const. - p[(º"'ox)2 + (ªcf,\2]
2 ºY J ,
(b) Show that there exist exactly two functions
u¡(x) = a 11 sin 1rx + a12 sin 21rx,
where cf, is the velocity potential. Derive the field equation governing cf, and the uix) = a 21 sin 1rx + a 22 sin 21rx,
corresponding natural boundary conditions according to Bateman's principie.
[H. Bateman, Proc. Roy. Soc. London, A, 125 (1929) 598-618.] notwithstanding a constant factor, which satisfy these necessary conditions
mentioned in (a).
10.6. Let (c) Show that for one of the functions mentioned in (b), say, u1(x), the inequality
n
T = .2 P µ/t, q
µ,v=l
1, .. . , qm) q /t) q.(t),
J[u1 ] < J[u]
U = U(t, q1 , . . . , qm), holds for all functions u(x) in D. Then show that for the other solution, uix),
and
L = T - U, the inequality
J[u] < J[u2]
where qµCt) = dq)dt, and q)t0 ) = aµ, q/t1 ) = bµ, (¡¡ = 1, ... , m); aµ , bµ are given
numbers. Derive the Euler differential equations for the variational problem holds for all functions u(x) in D .
connected with the functional (d) Does there exist any relation between J[u 1 ], J[u 2] and the eigenvalues of the
ti eigenvalue problem

Prove that
J[q¡, ... , qm] =
J to
L dt.
with
- u' (x) = J.u(x), 0 ,<X,< 1,

u(O) = u(l) = O.
d(T + U)
dt = o, 10.9. Clapeyron's theorem states that, if a linear elastic body is in equilibrium
under a given system of body forces F; and surface forces T ;: then the strain energy
if oU/ot = O, oPµJot = O, ¡¡, v = 1, ... , m . Hint: Use the Euler's relation for of deformation is equal to one-half the work that would be done by the external
homogeneous functions. forces (of the equilibrium state) acting through the displacements u; from the
10.7. Fiad the curves in the x,y-plane such that unstressed state to the state of equilibrium; i.e.,

Lb V2E - n2y2 ds, ds 2 = dx 2 + dy 2 f VF;u; dv + LT;u; dS f

= 2 VW dv.

is stationary, where E and n are constants and the integral is taken between fixed Demonstrate Clapeyron's Theorem for:
end points. (a) A simply supported beam under its own weight [Fig. P10.9(a)].
 312 ENERGY THEOREMS
Chap. 10 Chap. 10 PROBLEMS 313

(b) A rod under its own weight [Fig. PI0.9(b)], the following equilibrium equations in sorne average sense with respect to the

u(x) = gpL2[1
-
E
- - (x)2 - L-x] .
2 L
z-direction:
ca., car. ªrr - ªee p
-~-
"r
+ -~-
r,z
+ r = o,

~
ca,. ca.. ªr•
,JJ;--

(a)
~
If (b)

Pl0.9.
~b~
(el
-cr
+ - cz + -r = 0.

10.11. Consider a pin-ended column of length L and

bending stiffness El subjected to an end thrust P. A spring
is attached at the middle of the columnas shown in Fig. Pl 0.11.
When the column is straight the spring tension is zero. If the
column deflects by an amount 6., the spring exerts a force R on
(c) A strip of membrane of infinite length and width b [Fig. Pl0.9(c)J, under a the column :
constant pressure p 0, for which R = K6. - ot6.3, (K > O, ot > O).
Wo 3(1 - 112) Pob4
w = w0(1 - x 2 ), 3/ -4· Derive the equation of equilibrium of the system. It is per-
h 64 Eh missible to use the Euler-Bernoulli approximation for a beam, ?
10.10. The "poker chip" problem for which the strain energy per unit length is
(Max Williams). To obtain a nearly p
triaxial tension test environment for iEl · (curvature) 2
and Pl0.11
polymer rnaterials, a "poker chip" test bending moment = El· (curvature).
specimen (a short circular cylinder) is
10.12. For the column of Prob. 10.11 under the axial load P, is the solution
glued between two circular cylinders
(Fig. Pl0.10). When the cylinder is sub- E G v
unique? Under what situation is the solution nonunique? What are the possible
2 2 2 solutions when the uniqueness is lost?
jected to simple tension, the center of ' '
the poker chip is subjected to a triaxial 10.13. A linear elastic beam of bending stiffness El is supported at four equi-
tension stress field. Let the elastic con- ..l:h distant points ABCD, Fig. Pl0.13. The supports at A and D are pin-ended. At B
stants of the media be E 1 , G1 , 111 and E 2 , E1,G1,v1 r and C, the beam rests on two identical nonlinear pillars. The characteristic of the
G2, 11 2, as indicated in Fig. Pl0.10, with
E 1 <{; E 2 , 111 ~ 0.5. Assume cylindrical
T pillars may be described as "hardening" and is expressed by the equation
2o K6. = R - {JR 3 (K > O, fJ > O),
symmetry, and obtain approximate ex-
pressions of the stress field by the E2, G2,v2-+- where R is the reaction of the pillar, 6. is the downward deflection of the beam at the
following methods. point of attachment of the beam to the pillar, K and fJ are constants.
A load Pis applied to the beam at the mid-span point. f ind the reactions at the
(a) Use the complementary energy
theorem and a stress field satisfying H+ITTHT+ supports B and C. One of the two mínimum principies (of potential energy and of
complementary energy) is easier to apply to this problem. Solve the problem by a
stress boundary conditions and the Pl0.10 Poker chip specimen. variational method with the appropriate mínimum principie.
equations of equilibrium.
(b) Use the potential energy theorem and assumed displacements satisfying dis-
placement boundary conditions.
p
(c) Experimental results suggest that the following displacement field is reasonable:
z 8 e
W = Wo"i,_• - h < z.-;;; h,

u = (1 - ~)g(r),

where g(r) is as yet an unknown function of r . Obtain the governing equation

for g(r) and its appropriate solution for this problem by attempting to satisfy Pl0.13 Pl0.14
314 ENERGY THEOREMS Chap. 10

10.14. Consider a square plate loaded in the manner shown in Fig. PlQ.14.
Derive Euler's equation and the natural boundary conditions for V to be a minimulJI
when w(x, y ) is varied.
V = U + A, 11
D (1 (1
U = 2 Jo Jo [(w.,., + w1111) - 2(1 - v)(w.,.,w1111 - w.,; )] dx dy,
2

A =
i 1f1
00
qwdxdy -
11(
0
aw) /"'=1dy +
M ., -¡¡-
X X= O 11 /"'Z=O=1
O
(Q.,w) dy
HAMILTON'S PRINCIPLE,
WAVE PROPAGATION,
+
i1 l ow /"',11=1
=1 11 (
(H.,11 a) dy - ow) 111=1dx
M 11 -¡¡-

i 11=0 +il
0 Z=O O '.Y 11=0 APPLICATIONS OF GENERALIZED
+ (Q 11 w) dx
o 111=1dx.
(H.,11 / )
0 0 '.Y 11=0 COORDINATES

In dynamics, the counterpart of the minimum potential energy theorem is

Hamilton's principie. In this chapter we shall discuss this important principie
and its applications to vibrations and wave propagations in beams.
Toward the end of the chapter a brief discussion of the so-called direct
methods of solving variational problems is given. The basic idea is to apply
the concept of generalized coordinates to obtain approximate solutions for
a continuous system by reducing it to one with a finite number of degrees
of freedom. Severa! important methods-those of Euler, Rayleigh-Ritz-
Galerkin, and Kantrovich-will be outlined. The question of convergence,
however, must be referred to mathematical treatises. See Bibliography 11.3
on p. 498.

11.1. HAMILTON'S PRINCIPLE

For an oscillating body, with displacements ui so sm.all that the accelera-

tion is given by éJ2u;/ot 2 in Eulerian coordinates (Sec. 5.2) the equation of
small motion is

(1) <l¡¡.; + F; = p a2u;

at 2
•

where pis the density ofthe material and F; is the body force per unit volume.
Let us again consider virtual displacements bu; as specified in Sec. 10.7, but
instead of a body in static equilibrium we now consider a vibrating body.
The variations Ou; must vanish over the boundary surface S u, where values of
displacements are prescribed; but are arbitrary, tri pi y differentiable over the
domain V; and are arbitrary also over the rest of the boundary surface Su,
where surface tractions are prescribed.
315